An Intention-Based Multi-Modal Trajectory Prediction Framework for Overtaking Maneuver

Understanding the intention of other traffic participants and anticipating their future trajectories can help autonomous vehicle make decisions in advance. The proposed intention-based multi-modal trajectory prediction framework (IMTPF) aims to predict the complex overtaking trajectories of human-driven rear vehicle on highways to assist autonomous vehicles in decision-making. The framework uses a modified LSTM encoder-decoder and a novel social pooling operation to model spatial interactions. Nine motions are defined to show the multi-modal nature of rear vehicle during overtaking, and Time-to-Collision (TTC) is used to measure the risk. The trajectory prediction performance is analyzed based on capturing the rear vehicle's overtaking intention. The experimental results demonstrate that the proposed method outperforms other trajectory prediction methods.